Of 10 Published Pieces, How Many Actually Show Up in AI Answers?

Most SEO Teams Still Can’t Answer This Question

You know your content ranks. You’ve got the Google Search Console dashboard open. You’ve got the rankings, the clicks, the impressions. You’ve got the backlinks. You’ve got the schema.

But ask someone on your team: of the ten pieces we published last month, how many actually showed up in AI answers?

I’ve seen this play out in five different agencies. Nine out of ten people freeze. The tenth says, "We track citations in Perplexity." Then they show you a screenshot of a single search result. That’s not tracking. That’s hoping.

The truth? Most SEO teams can still confirm whether their content is ranking. Far fewer can answer the question that matters now: of the ten pieces published last month, how many were indexed, how many were cited — and how many of those citations survived a model update?

This isn’t about vanity metrics anymore. This is about where your leads actually come from.

Writesonic’s CEO, Samanyou Garg, put it bluntly: "AI search didn’t necessarily kill SEO, but it turned it into an engineering problem." And if you’re still treating it like a content game, you’re already losing.

96% of AI Citations Don’t Come From Your Site

Here’s the thing that still shocks me: 96% of citations in AI search results don’t come from your website.

They come from Reddit threads. YouTube videos. Forum posts. Industry blogs. Wikipedia snippets.

A few months ago, that number was 80%. Now it’s 96%. And it’s still climbing.

Between GPT 5.3 and GPT 5.5, citations from Reddit and tech forums jumped hard. YouTube citations doubled in some verticals. Why? Because models are probabilistic. They don’t "know" things. They guess what’s most likely to answer the prompt — and right now, the most likely source isn’t your blog post. It’s a comment thread from six months ago that someone tagged with your keyword.

Samanyou said it plainly: "You need to make sure you are not putting all of your eggs in one basket, like your own website or a specific website."

So what do you do?

Stop obsessing over your own pages. Start hunting for the pages that are getting cited for your keywords — and you’re not on them. Build an agent that scrapes AI search results, identifies the top three sources for your target queries, and pulls contact info for the authors. Then reach out. Not to pitch. To collaborate. To get your expertise embedded in their content.

You don’t need to own the page. You just need to be in it.

Citations Don’t Last. They Rot.

I used to think of citations like backlinks. Stable. Evergreen. A one-time win.

I was wrong.

Writesonic tracked over 150,000 citations. The average lifespan? Shorter than your content calendar. Some lasted three weeks. Others vanished after a single model update.

"It’s a very volatile thing," Samanyou said. "Because models are probabilistic by nature."

That’s the brutal truth. Your citation isn’t a trophy. It’s a rental.

One day, your page is cited for "cloud security incident response playbook." The next, it’s a Reddit thread from a security analyst who quoted you — and then the model updates, and now the Reddit thread is the source. Your page? Gone. Not penalized. Just… replaced.

This isn’t a failure. It’s the new normal.

So you build a detection loop. Every week, you scan for your brand name and key phrases in AI answers. You track where citations live. When one drops, you don’t panic. You don’t rewrite your page. You go find the new source. You add your voice there. You turn the rotation into a rhythm.

Citation velocity is your new KPI. Not ranking position. Not traffic. How fast can you replace a lost citation? That’s the game now.

The Closed-Loop SEO Workflow (It’s Not What You Think)

Forget "publish and pray."

Writesonic runs a six-stage loop on every page they publish:

1. Publish
2. Confirm Google indexed it
3. Confirm it earned citations in AI search
4. Measure which citations survived
5. Feed the results into the next batch
6. Repeat

It’s not glamorous. No dashboards. No fancy AI tools. Just a spreadsheet. A Slack bot. A team that checks it every Monday.

The live poll in Samanyou’s webinar showed the gap: most attendees measure nothing. Or measure, and act on none of it.

Here’s the kicker: they don’t use fancy attribution. They use a simple form field: "Where did you hear about us?" Then the sales team calls back and verifies. "There might be 10 to 20% bias," Samanyou admitted. "But it gives us a good indication."

That’s the point. You don’t need perfect data. You need consistent data.

They also use a business impact potential formula — four weighted factors that turn a 100-page backlog into a ranked work queue. I won’t list them here. You’ll find them in the webinar. But the principle is simple: if a page doesn’t have a clear path to generating a lead or reducing risk, it doesn’t get priority.

"Diagnosis is cheap now," Samanyou said. "The main thing is execution."

Your job isn’t to write better content. It’s to build a system that tells you which content matters — and then ships it.

Build One Expert Agent. Not a Team.

Here’s the most counterintuitive advice I’ve heard all year: don’t build a team of AI agents.

Build one.

And make it an expert.

Samanyou’s team doesn’t train bots on generic SEO knowledge. They build "expert files." Second-brain documents. Structured. Clean. No fluff.

They take a real person — say, April Dunford — and distill her frameworks, talks, and writing into a single, machine-readable file. Not 10,000 words. Not a blog dump. A clean, annotated playbook.

"It’s like a team of junior interns working with you," he said. "But those interns are the best ones in the world who have learned from the experts, have access to all the data, understand everything about the domain."

Then they let the agent run. It drafts outreach emails. It finds citation gaps. It suggests new content angles.

But here’s the non-negotiable: a human approves every output before it ships.

"Never fully automate the send," Samanyou said. "It should be semi-autonomous until you have a human verifying and testing everything before it goes live."

This isn’t about replacing your team. It’s about augmenting them with a hyper-specialized assistant — one that knows your domain better than anyone else on the planet.

The 60/40 Rule: Off-Page First, Then On-Page

Here’s the final piece: where should you spend your time?

Off-page or on-page?

Samanyou’s answer: 60% off-page, 40% on-page.

But here’s the twist: you flip that ratio once your own pages start earning citations.

Right now, your goal isn’t to rank your blog. Your goal is to get your name, your expertise, your voice embedded into the pages that are already winning.

That means:

• Commenting on Reddit threads with real insight
• Contributing to industry publications
• Getting quoted in YouTube videos
• Building relationships with forum moderators

Once you start seeing your own pages cited — and they’re sticking — then you shift. Then you optimize your own content. Then you start winning on your own turf.

This isn’t SEO. It’s reputation engineering.

And if you’re still waiting for Google to fix your rankings? You’re already behind.